Probability Forecasting for Inflation Warnings from the Federal Reserve

• Activity Number: 412
• Type: Topic Contributed
• Date/Time: Tuesday, August 6, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Business and Economic Statistics Section
• Author(s): Shaun Vahey*+ and Anthony Garratt and James Mitchell
• Companies: ANU and Birkbeck, University of Lodon and Warwick University
• Keywords: Vector autoregression ; Forecasting inflation ; Probability forecasts ; Cost-Loss ratio

Abstract

Forecasting Macroeconomic Trends Gian Luigi Mazzi 208871

Forecasting inflation constitutes a primary task of monetary policymakers. The US Federal Reserve and other central banks frequently publish warnings about extreme inflation events. In this paper, we propose a probabilistic approach to forecasting inflation specifically for the purpose of issuing warnings about inflation events to the public. We combine the predictive densities produced from Vector Autoregressive (VAR) models utilizing prediction pools and evaluate our ensemble probabilistic forecasts for quarterly US data from 1990:1 to 2012:1. Adopting a cost-loss ratio for forecast evaluation indicates that economic loss could be reduced by up to 45 percent, relative to the benchmark model in which inflation follows an integrated moving average process. A `climatological' forecast (based on the unconditional probability of the inflation event) also typically outperforms the benchmark, but fails to match the ensemble VAR. Our finding indicates considerable scope for using formal forecasting methods to forewarn the public of extreme inflation events.